With the deepening of electricity market reform, the diversity of users’ electricity demand and the complexity of electricity retail packages have put forward higher requirements for package design and recommendation technology. In this paper, we propose a three-stage collaborative approach that integrates the modeling of users’ electricity consumption behavior, the global optimization of differential evolution (DE) algorithm and the intelligent recommendation of attention factor decomposition machine (AFM), aiming at achieving the dynamic design and accurate recommendation of electricity retail packages. An extensible package product family GBOM is constructed based on the quintuple information expression model and modular design rules, and the differential evolution (DE) algorithm is used to efficiently search for the optimal bidding strategy in the highdimensional solution space, combined with the constraint rules to adaptively deal with the complex coupling relationship between modules. An intelligent recommendation algorithm based on AFM is further proposed to enhance the model’s sensitivity to the key features of users’ electricity behavior by introducing an attention mechanism to dynamically assign feature cross weights. The experimental results show that the differentiated pricing strategy has a significant effect in peak and valley time regulation, and the peak time price is 45.4% higher than the fixed price (e.g., the peak time price of user 1 is 453.28 yuan/MWh), and the valley time price is reduced by 47.3% (the valley time price of user 1 is 164.23 yuan/MWh).The AFM recommendation algorithm combines the user’s load characteristics with the monthly electricity consumption grading, and the recommendation accuracy rate is reaches 94.09% (when the number of historical purchases Te=7), which is a significant improvement over traditional methods (e.g., 89.41% accuracy rate of the mean weight method). Through comparative analysis, the DE-AFM method performs optimally in terms of balancing error and practicality (RMSE=0.0144, accuracy rate 94.09%), which verifies its stability and accuracy in complex scenarios.
